Although a large amount of research has been done in speech animation, both 2D and 3D, one shortcoming of current speech animation methods is that they cannot generate dynamic expressions automatically. In this paper, an automatic technique for synthesizing novel dynamic expression for 3D speech animation is presented. After a Phoneme-Independent Expression Eigen-Space (PIEES) is extracted from expressive facial motion capture data by Principal Component Analysis (PCA), texture synthesis methods are used to synthesize novel dynamic expression from the PIEES. Control for expression intensity dynamics is also provided for animators via expression intensity curves.